Object Hallucination Probing on GQA POPE Random
89.93Accuracy (GQA POPE)Scalpel
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ScalpelBase Model=LLaVA-1.5-7B2026.02 | 89.93 | — | — | — | — | — | — | — | — | — | 89.87 | — | — | |
| MESAModel=LLaVA-v1.52026.04 | 89.5 | — | — | — | — | — | — | — | — | — | 89.45 | 90 | 89 | |
| ICTBase Model=LLaVA-1.5-7B2026.02 | 89.3 | — | — | — | — | — | — | — | — | — | 89.49 | — | — | |
| ICTModel=LLaVA-v1.52026.04 | 89.26 | — | — | — | — | — | — | — | — | — | 88.78 | 92.93 | 85 | |
| CoFi-DecBackbone=LLaVA-1.52025.12 | 89.03 | — | — | — | — | — | — | — | — | — | 89.06 | 81.1 | — | |
| VAFModel=LLaVA-v1.52026.04 | 89 | — | — | — | — | — | — | — | — | — | 88.63 | 91.66 | 85.8 | |
| VTIModel=LLaVA-v1.52026.04 | 88.37 | — | — | — | — | — | — | — | — | — | 88.02 | 90.73 | 85.47 | |
| OPERABase Model=LLaVA-1.5-7B2026.02 | 88.13 | — | — | — | — | — | — | — | — | — | 88.91 | — | — | |
| MESAModel=Qwen-VL2026.04 | 87.97 | — | — | — | — | — | — | — | — | — | 86.67 | 97.11 | 78.27 | |
| NulluModel=LLaVA-v1.52026.04 | 87.7 | — | — | — | — | — | — | — | — | — | 88 | 85.9 | 90.2 | |
| CoFi-DecBackbone=Qwen-VL2025.12 | 87.14 | — | — | — | — | — | — | — | — | — | 86.32 | 94.65 | — | |
| DeGFBackbone=LLaVA-1.52025.12 | 87.09 | — | — | — | — | — | — | — | — | — | 87.96 | 80.46 | — | |
| CoFi-DecBackbone=InstructBLIP2025.12 | 86.78 | — | — | — | — | — | — | — | — | — | 86.39 | 87.06 | — | |
| VCDBase Model=LLaVA-1.5-7B2026.02 | 86.65 | — | — | — | — | — | — | — | — | — | 86.99 | — | — | |
| VCDModel=LLaVA-v1.52026.04 | 86.65 | — | — | — | — | — | — | — | — | — | 86.99 | 84.85 | 89.24 | |
| ICDModel=Qwen-VL2026.04 | 86.2 | — | — | — | — | — | — | — | — | — | 84.62 | 95.55 | 75.93 | |
| VAFModel=Qwen-VL2026.04 | 86.11 | — | — | — | — | — | — | — | — | — | 86.11 | 96.87 | 77.53 | |
| RITUALBackbone=LLaVA-1.52025.12 | 86.1 | — | — | — | — | — | — | — | — | — | 87.31 | 80.3 | — | |
| RITUALBackbone=Qwen-VL2025.12 | 86.1 | — | — | — | — | — | — | — | — | — | 84.81 | 93.78 | — | |
| DeGFBackbone=Qwen-VL2025.12 | 85.95 | — | — | — | — | — | — | — | — | — | 85.08 | 94.22 | — | |
| ICDModel=LLaVA-v1.52026.04 | 85.9 | — | — | — | — | — | — | — | — | — | 85.29 | 89.1 | 81.8 | |
| M3IDBackbone=Qwen-VL2025.12 | 85.69 | — | — | — | — | — | — | — | — | — | 84.67 | 93.11 | — | |
| VCDModel=Qwen-VL2026.04 | 85.59 | — | — | — | — | — | — | — | — | — | 85.33 | 86.88 | 83.84 | |
| DeGFBackbone=InstructBLIP2025.12 | 85.4 | — | — | — | — | — | — | — | — | — | 85.12 | 85.64 | — | |
| ICTModel=Qwen-VL2026.04 | 85.33 | — | — | — | — | — | — | — | — | — | 86.12 | 95.95 | 78.13 | |
| VCDBackbone=Qwen-VL2025.12 | 85.21 | — | — | — | — | — | — | — | — | — | 84.21 | 92.05 | — | |
| VanillaModel=LLaVA-v1.52026.04 | 84.96 | — | — | — | — | — | — | — | — | — | 86.25 | 79.44 | 94.33 | |
| RegularBackbone=Qwen-VL2025.12 | 84.9 | — | — | — | — | — | — | — | — | — | 83.96 | 89.51 | — | |
| RITUALBackbone=InstructBLIP2025.12 | 84.87 | — | — | — | — | — | — | — | — | — | 85.39 | 82.52 | — | |
| VTIModel=Qwen-VL2026.04 | 84.33 | — | — | — | — | — | — | — | — | — | 81.94 | 96.73 | 71.07 | |
| M3IDBackbone=LLaVA-1.52025.12 | 84.2 | — | — | — | — | — | — | — | — | — | 85.77 | 78 | — | |
| VanillaBase Model=LLaVA-1.5-7B2026.02 | 83.73 | — | — | — | — | — | — | — | — | — | 82.95 | — | — | |
| VCDBackbone=LLaVA-1.52025.12 | 83.23 | — | — | — | — | — | — | — | — | — | 85.05 | 76.73 | — | |
| M3IDBackbone=InstructBLIP2025.12 | 83.07 | — | — | — | — | — | — | — | — | — | 83.87 | 80.06 | — | |
| VCDBackbone=InstructBLIP2025.12 | 82.83 | — | — | — | — | — | — | — | — | — | 83.56 | 80.16 | — | |
| RegularBackbone=LLaVA-1.52025.12 | 82.23 | — | — | — | — | — | — | — | — | — | 84.03 | 76.32 | — | |
| RegularBackbone=InstructBLIP2025.12 | 79.67 | — | — | — | — | — | — | — | — | — | 80.99 | 76.05 | — | |
| VanillaModel=Qwen-VL2026.04 | 74.76 | — | — | — | — | — | — | — | — | — | 66.6 | 98.43 | 50.33 | |
| NulluModel=Qwen-VL2026.04 | 74.73 | — | — | — | — | — | — | — | — | — | 65.55 | 98.43 | 50.26 | |
| VCDModel=Qwen-VL, Decoding=VCD2026.01 | 59.7 | — | — | — | — | — | — | — | — | — | 56.43 | 61.41 | 52.2 | |
| SDCDModel=Qwen-VL, Decoding=SDCD2026.01 | 58.97 | — | — | — | — | — | — | — | — | — | 55.35 | 60.7 | 50.87 | |
| Regular DecodingModel=Qwen-VL, Decoding=Regular2026.01 | 58.93 | — | — | — | — | — | — | — | — | — | 54.44 | 61.13 | 49.07 | |
| BaselineBackbone=LLaVA-1.52025.06 | — | 89.47 | 77.03 | 63.83 | 70.23 | 74.87 | 68.07 | — | — | 70.81 | — | — | — | |
| BaselineBackbone=Qwen-VL-Chat2025.06 | — | 87.23 | 82.53 | 73.03 | 74 | — | — | 65.1 | 74.3 | 73.95 | — | — | — | |
| CLAIMBackbone=LLaVA-1.52025.06 | — | 84.27 | 85.13 | 83.57 | 82.53 | 82.67 | — | — | — | 83.63 | — | — | — | |
| CLAIMBackbone=Qwen-VL-Chat2025.06 | — | 85.17 | 79.93 | 81.9 | 62.03 | — | — | 80.4 | — | 77.89 | — | — | — | |
| VCDBackbone=LLaVA-1.52025.06 | — | — | 78 | 68.33 | 71.47 | 73.63 | 70.2 | — | — | 72.33 | — | — | — | |
| VCDBackbone=Qwen-VL-Chat2025.06 | — | — | 83.63 | 76.1 | 73.43 | — | — | 55.5 | 80.2 | 73.77 | — | — | — |